Extraction of robust functional connectivity patterns across psychiatric disorders using principal component...

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Biomarkers for psychiatric disorders are researched using resting-state functional connectivity (FC).
• Identifying abnormal FCs is crucial for therapeutic targets, beyond just diagnostic performance.
• Previous methods often overlooked the effect size and robustness of selected FCs.

Purpose of the Study:

• To robustly identify and extract abnormal functional connectivity (FC) patterns in major depressive disorder (MDD), schizophrenia (SCZ), and autism spectrum disorder (ASD).
• To compare the effectiveness of a PCA-based feature selection method against conventional supervised methods for FC biomarker extraction.
• To validate the identified FCs for their potential as therapeutic targets.

Main Methods:

• Utilized approximately 5,000 resting-state functional magnetic resonance imaging (rs-fMRI) datasets across six cohorts.
• Employed a Principal Component Analysis (PCA)-based feature selection method.
• Compared PCA-based selection with conventional supervised feature selection techniques.

Main Results:

• The PCA-based method robustly extracted FCs with significantly larger effect sizes across MDD, SCZ, and ASD compared to supervised methods.
• Identified 78 essential FCs for MDD, 69 for SCZ, and 81 for ASD.
• Discovered that these essential FCs predominantly involve thalamic and motor network connections.

Conclusions:

• A PCA-based feature selection approach robustly identifies abnormal functional connectivity (FC) in psychiatric disorders consistently across datasets.
• The identified FCs demonstrate larger effect sizes and greater robustness, making them suitable for neurofeedback or neuromodulation therapies.
• This research contributes to understanding the neural mechanisms underlying psychiatric disorders by highlighting specific abnormal brain signatures.